The sample_rate common dimension was set once at service creation
time and never updated when SetSampleRate was called later. This
caused commands like 'gh skill publish' that override the sample
rate via PersistentPreRunE to report the wrong sample_rate in
telemetry events (e.g. 1 instead of 100).
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
When installing a skill whose SKILL.md contains github-repo metadata
pointing to a different repository, the CLI detects it as a re-published
skill and offers to redirect the install to the upstream source.
In interactive mode, the user is prompted to choose between the
re-publisher (default) and the upstream. In non-interactive mode,
the install proceeds from the re-publisher with a notice.
The --upstream flag skips the prompt and redirects to upstream directly,
enabling non-interactive upstream installs in CI/scripts.
If the user chooses upstream, the install restarts from that repo,
resolving the latest version and discovering skills fresh. A
skill_upstream_redirect telemetry event is emitted to track redirects.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Use a two-pass search so exact DisplayName and Name matches are
preferred over InstallName. This avoids incorrectly selecting a
plugins-convention skill via InstallName when a standard namespaced
skill with a matching DisplayName exists later in the sorted slice.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
The preview command's selectSkill function only matched skills by
DisplayName() and Name. For plugins-convention skills, the install
hint outputs InstallName() (namespace/name), which matched neither -
DisplayName() includes a [plugins] prefix and Name is just the base
name. This caused 'skill not found' errors when users ran the
suggested preview command after install.
Add InstallName() as an additional match criterion so that namespaced
skill identifiers produced by the install hint are accepted.
Closes#13248
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Addresses review feedback on PR #13244:
- Flag that --template/-T collides with body-template flags on
pr create / issue create, so agents should check --help.
- Recommend the single quoted-string form for gh search (matching
--search), since multi-arg invocations join oddly and can produce
invalid queries.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Avoids drift between this skill and the actual validation logic in
publish. Agents should run `gh skill publish --dry-run` to learn the
current rules from the source of truth.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Relax skill discovery to recognize skills/ directories at any depth
in the repository tree, not just at the root. This enables repos like
hashicorp/agent-skills that organize skills under prefixes such as
terraform/code-generation/skills/*/SKILL.md.
Changes:
- matchSkillConventions: add checks for <prefix>/skills/<name>/SKILL.md
and <prefix>/skills/<ns>/<name>/SKILL.md at any depth
- isSkillPath: also match paths containing /skills/ for direct
path-based install
- DiscoverSkillByPath: fix namespace detection to find the skills
segment anywhere in the path
- Update error messages and help text to mention nested conventions
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
- --fix and --dry-run now error when combined
- "Ready to publish!" is suppressed when --fix is set since user must
commit first
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
When --fix is used, return early after applying fixes instead of
continuing to the publish flow. The fixed files need to be committed
before publishing, so proceeding would fail anyway since the repo
would not be in sync.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Move --allow-hidden-dirs filtering logic from the discovery package to
the install command, addressing review feedback. Discovery functions now
always return all skills (including hidden-dir), and callers decide how
to handle them.
Changes:
- DiscoverSkillsWithOptions/DiscoverLocalSkillsWithOptions always return
hidden-dir skills; callers filter using IsHiddenDirConvention()
- DiscoverSkills/DiscoverLocalSkills (convenience wrappers) auto-filter
hidden-dir skills for backward compatibility with preview/update/publish
- Remove --allow-hidden-dirs reference from discovery error messages
- Add filterHiddenDirSkills in install.go with caller-side flag logic
- Inline warning using heredoc.Docf, remove printHiddenDirWarning
- Add inline comments in matchHiddenDirConventions (babakks nitpicks)
- Add non-hidden-namespaced dir and no-skills-at-all test cases
- Add --allow-hidden-dirs tests in TestNewCmdInstall, TestInstallRun,
and TestRunLocalInstall
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
* add support for installation in multiple agent host
* print correct dir in warning
* remove dir as it depends on user vs project installation scope
* Move comment closer to assertion in registry test
Move the explanatory comment from above the map initialization to
directly above the assertions it describes, per review feedback.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
* List supported agent names and IDs in help text
Replace the self-referencing "run --help" sentence with an inline list
of all supported --agent values showing Name (id) pairs.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
* Use heredoc.Docf for Kiro CLI post-install hint
Replace individual fmt.Fprintln calls with a single heredoc.Docf block
for the Kiro CLI post-install guidance, per review feedback. Also
shorten the --agent flag usage line by overriding the auto-generated
enum list with a reference to the supported values in the help text.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
---------
Co-authored-by: Kynan Ware <47394200+BagToad@users.noreply.github.com>
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
* Add skills specific telemetry
* Remove VisibilityFuture, inline goroutine at call sites
The VisibilityFuture/FetchRepoVisibilityAsync/Wait wrapper was an
unidiomatic async abstraction built for a single pattern used in
exactly two call sites. In Go the channel is already the future;
wrapping it in a struct with a Wait(timeout) method adds no value.
Delete the abstraction and inline a local visResult struct,
buffered channel, goroutine, and select at each call site. Behavior
is preserved exactly: err -> "unknown", timeout -> "unknown",
success+public -> include skill_names.
FetchRepoVisibility (synchronous) is kept as-is.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
* Fix nonsense copilot tests
* Update telemetry tests for public-only dims and search event removal
Production telemetry emission changed:
- preview: skill_owner/skill_repo/skill_name (renamed from skill_names)
are now emitted only when repo_visibility=public.
- install: skill_owner/skill_repo/skill_names are now emitted only
when repo_visibility=public.
- search: the initial skill_search event was removed entirely; the
skill_search_install event no longer carries query/owner dims.
Update tests to match: rename skill_names -> skill_name in preview,
make owner/repo assertions conditional on public visibility in both
preview and install, and reduce the search test to a single event
with explicit Empty assertions for the removed query/owner dims so a
privacy regression cannot pass silently.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
* Test CategorizeHost and switch telemetry to skill_host_type
Add TestCategorizeHost covering all four classification branches
(github.com, ghes, tenancy, uncategorized) with cases verified
against the real ghauth implementation rather than guessed.
Update install and preview unit tests to assert the new
skill_host_type dimension name, and fix a typo in the preview
acceptance txtar (skill_hos_type -> skill_host_type).
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
* Shrink visibility wait and test unknown visibility
The 2s visibilityWaitTimeout was wildly overprovisioned: by the time
telemetry emission reaches the select, the command has already done
several serial GitHub REST calls (and for install, a git sparse-checkout
plus possibly interactive prompts), so the one-call visibility fetch
has almost always completed. Drop the timeout to 200ms — a short safety
net for a stalled REST call, not a wait budget for a healthy one.
Also adds a table-driven case to TestFetchRepoVisibility covering an
unknown/future visibility value from the API, addressing @babakks'
review nitpick.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
---------
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
